Rpart(decision tree) not choosing attribute with highest information gain as root node

Question

i computed manually every information gain of my attributes and cross checked with FSelector which gives the same results. However when i made the classifier in rplot, it ch